By the way, there is a very interesting tournament going on currently -- TCEC, unofficial computer chess championship [1]. The main intrigue of the tournament is the fight between Stockfish and Komodo. Houdini, dominating engine in the last few years, already lost any chances to go into the superfinal -- two very strong challengers, Komodo and Stockfish, in the last few months were able to eat huge advantage of Houdini. The author of Komodo, Don Dailey, is terminally ill, according to his wife he has just a few days to live, so it's possible that he will not know if his engine will win or not. It's really heartbreaking, he was very active in the chat of TCEC just a few days ago, commenting games, discussing computer chess and so on. Stockfish is an open-source engine that uses distributed testing framework, so anybody can participate in its development donating CPU time [2]. Both engines are very strong and have relatively equal strength.
Wouldn't a match with an open-source engine be unfair? The competing close-source engine could embed the opponents source inside it to predict the exact moves that the opponent would make, thus gaining a speed advantage.
Even more interestingly, what would a match between two open-source engines look like, if they both embedded each other?
This is a very interesting tournament. AFAIK it is the only computer tournament with long time controls and equal hardware for both sides. I have been following it for some years already since the times of Old TCEC. The most interesting games annotated by GM or FM level players can be found here [1]. My favourite is the game Old TCEC Elite Match - Season 1 game 1.1, in which Houdini crushed Rybka with black pieces very convincingly by sacrificing 3 pawns to get a devastating advantage.
Do you know of a document explaining the most important ideas in these chess engines? I know about alphabeta search and the like, but I'm wondering what the secret sauces in the recent best chess engines are.
Here you go: http://chessprogramming.wikispaces.com/ It's an ultimate source about chess programming. Also, Stockfish code is short (several KLOC) and well commented.
To make it truly interesting (from the algorithmic point of view), these competitions should be limited to a standard machine with no connectivity. Otherwise it's an arms race.
In TCEC all engines run on the same 2x 8-core Xeon. It's WCCC that allows engines to use arbitrary hardware -- e.g. on WCCC 2013 Jonny run on 2400-core cluster, but lost anyway to Junior that used 24-core Xeon.
Right now the chess produced is still more interesting than the algorithms themselves. Once we've hit diminishing returns there, it's likely that they'll realign as you've suggested.
According to CCRL [1], Komodo 6 and Stockfish 4 are 20-25 Elo weaker than Houdini 3 (in the last few years Rybka wasn't able to compete with Houdini, in TCEC it wasn't even able to qualify into Stage 4). However TCEC uses development versions of Komodo and Stockfish that are much stronger than their latest stable releases.
It's hard to say precisely because top players don't really play top computers competitively, but the best computer engines appear to be well above 3000 by now. The estimates on the TCEC page put Stockfish and Komodo both at around 3100.
[1] http://tcec.chessdom.com/ [2] http://tests.stockfishchess.org/